The essence of aligning large LLM models is basically to conform to the existing social order. ChatGPT respects China's order, while China's open-source models, although reluctantly, also tolerate the order in the US. Maybe we can build a revisionist SFT dataset.
Any conversation
ะฆะฒะตัะพะบ ัะพะทั ะฝะต ะธะทะดะฐะตั ะณะพะปะพัะฐ, ะฐ ะฑะปะฐะณะพัั
ะฐะฝะธะต ะตะณะพ ะดะฐะปะตะบะพ ัะฐัั
ะพะดะธััั ะผะพะปัะฐ; ัะฐะบ ะฝะฐะดะพ ะถะธัั ะธ ะฒัะตะผ ั
ัะธััะธะฐะฝะฐะผ.
โะกะฒััะธัะตะปั ะคะตะพัะฐะฝ ะะฐัะฒะพัะฝะธะบ
The rose flower makes no sound, but its fragrance spreads far and wide in silence; so should all Christians live.
โSt. Theophan the Recluse
The government welds your door shut and then outlaws starting a business and living in McMansions, cars, skyscrapers, and tiny homes.
The governments bans body odor and then closes stores and bans delivery services.
The state closes bathrooms and then outlaws public urination.
The government closes the border and then arrests tourists for overstaying their visas.


